Endowment heterogeneity, incomplete information & institutional choice in public good experiments

Endowment heterogeneity, incomplete information & institutional choice in public good experiments

Endowment Heterogeneity, Incomplete Information & Institutional Choice in Public Good Experiments Journal Pre-proof Endowment Heterogeneity, Incompl...

2MB Sizes 0 Downloads 45 Views

Endowment Heterogeneity, Incomplete Information & Institutional Choice in Public Good Experiments

Journal Pre-proof

Endowment Heterogeneity, Incomplete Information & Institutional Choice in Public Good Experiments Lawrence R. De Geest, David C. Kingsley PII: DOI: Reference:

S2214-8043(19)30042-4 https://doi.org/10.1016/j.socec.2019.101478 JBEE 101478

To appear in:

Journal of Behavioral and Experimental Economics

Received date: Revised date: Accepted date:

21 January 2019 20 September 2019 7 October 2019

Please cite this article as: Lawrence R. De Geest, David C. Kingsley, Endowment Heterogeneity, Incomplete Information & Institutional Choice in Public Good Experiments, Journal of Behavioral and Experimental Economics (2019), doi: https://doi.org/10.1016/j.socec.2019.101478

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier Inc.

1

2

3 4

5 6

7

8 9

Highlights • We vary of the observability of heterogenous endowments in a public good experiment. • Groups get experience with central authority and peer punishment before voting which one to implement. • In Unobserved subjects with lower (higher) endowments prefer central authority (peer punishment). • Cooperating low endowment members receive more punishment in Unobserved. • Subjects with higher endowments in Unobserved undermine targeted sanctions by disguising contributions.

1

Endowment Heterogeneity, Incomplete Information & Institutional Choice in Public Good Experiments

1 2

Lawrence R. De Geest∗, David C. Kingsley∗∗

3

4

Abstract We study centralized and decentralized enforcement in social dilemmas with income inequality and incomplete information. Subjects are randomly assigned different endowments and, across treatments, these endowments can either be observed or not. After gaining experience with peer punishment and a simple central authority, groups voted on their preferred enforcement institution. Under complete information (endowments observed), most groups voted for peer punishment. Under incomplete information (endowments unobserved), most groups voted for central authority, and results suggest this preference was largely driven by subjects with lower incomes. Since free-riding could not be targeted when incomes were not observed, subjects with larger incomes tended to under-contribute, encouraging groups to self-impose central authority.

5 6 7

Keywords: Public goods, peer punishment, central authority, cooperation, experiment, institutions JEL: C92, D02, H41

Corresponding author. Department of Economics, Suffolk University. Thank you to Alex Smith, Arjun Sengupta and participants at the Appalachian Experimental and Environmental Economics Workshop and the New England Experimental Economics Workshop for helpful comments. Funding from the International Foundation for Research in Experimental Economics and the University of Massachusetts Lowell is gratefully acknowledged. ∗∗ Department of Economics, University of Massachusetts Lowell. Email address: [email protected] (Lawrence R. De Geest) ∗

Preprint submitted to Journal of Behavioral and Experimental Economics

October 7, 2019

1

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

1. Introduction Cooperation in social dilemmas requires well-designed institutions that spur individuals to act in the common interest. One of the most straightforward ways to promote cooperation is to punish selfishness with monetary sanctions (Ostrom et al., 1992; Fehr and Gächter, 2000; Chaudhuri, 2011). But who should levy these sanctions? Enforcement can be meted out by the individuals themselves through peer punishment or by a central authority. While both institutions provide similar benefits – they increase the private cost of acting selfishly – they impose different social costs. Peer punishment can be low cost when a credible threat is established and sanctions need not be used, or it can be high cost if anti-social or retaliatory punishment breaks out.1 On the other hand, enforcement by a central authority necessarily entails a fixed cost, even under full cooperation.2 For example, in the absence of speeding or crime in general, the police and the court system that supports them must still be paid for. It is thus of interest to determine when, and under what conditions, central authorities are preferred over peer punishment. Nicklisch et al. (2016) suggest that imperfect information may explain why modern society is largely characterized by central authorities. When individuals do not perfectly observe each other’s behavior because of some noise beyond their control, peer punishment becomes less effective. Specifically, imperfect information makes it difficult to target freeriders, thus limiting the costs of free-riding. At the same time, the accidental punishment of cooperators reduces the benefit of cooperation and risks sparking retaliatory punishment. Beyond the research of Nicklisch et al. (2016), which we return to below, this intuition is supported in two papers which investigate the effect of imperfect information within peer punishment. Both Grechenig et al. (2010) and Ambrus and Greiner (2012) report that introducing noise into peer punishment mechanisms significantly lowers net earnings relative to conditions without the opportunity to punish. Briefly, they observe that more, but less severe, punishment is deployed. The effect of this punishment, relative to the effect observed under perfect information, is that it attenuates the increase in contributions from freeriders who are punished 1

Peer punishment has been shown to improve welfare, but it is not uncommon to observe no increase or even a decrease in group welfare when peer punishment is relatively weak, expensive to deploy, or when there is anti-social/retaliatory punishment (Ostrom et al., 1992; Fehr and Gächter, 2000, 2002; Bochet et al., 2006; Sefton et al., 2007; Egas and Riedl, 2008; Gächter and Herrmann, 2008; Nikiforakis, 2008; Kosfeld et al., 2009). 2 Deterrent central authorities alter one’s incentives such that they are better off cooperating. See Tyran and Feld (2006); Galbiati and Vertova (2008); Kosfeld et al. (2009); Putterman et al. (2011); Markussen et al. (2014); Kamei et al. (2015) for detailed discussions concerning formal, centralized, institutions within social dilemmas.

1

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

and exacerbates the decrease in contributions from cooperators who are punished.3 Nicklisch et al. (2016) investigate a similar peer punishment regime with noise but allow group members to self-select into a centralized punishment or a no punishment regime. The centralized punishment regime effectively delegated the peer punishment mechanism to a single member of the group. Their results suggest that as the amount of noise increases, the preference for peer punishment decreases. However, this does not necessarily imply a preference for centralized punishment as an anti-dote to imperfect information. The preference for the centralized punishment regime in the noisy treatments is sensitive to the behavior of the delegated authority. Centralized punishment is preferred in the noisy treatments when the delegated authorities do better at targeting freeriders and rewarding cooperators.4 The literature thus, intuitively, suggests that preferences across institutions designed to solve social dilemmas are sensitive to the relative effectiveness of those institutions. In turn, the relative effectiveness across institutions appears to be, at least partially, explained by imperfect information. It follows that institutional preferences might also be tied to incomplete information, the case when individuals do not perfectly observe certain characteristics about others (Harsanyi, 1967). One way incomplete information could come about in social dilemmas is if income (or generally, one’s available resources) is heterogeneous but unobserved, giving individuals the opportunity to choose contributions that do not reveal their relative income and thus their capacity to contribute. In a town where residents have different incomes, raising funds for a local public good (say, resources for a library, or a park cleanup) could be done through voluntary contributions and some form of peer-to-peer enforcement (e.g. social punishment like ostracism) – or the the town could raise the funds through taxes. We can also think of more general examples of endowment heterogeneity. For instance, a firm hands a project to a team of employees, and each employee has a different “time endowment.” Employees with fewer commitments have more time to contribute to the project. But if they prefer to shirk, they can claim they are too busy to help. So, the firm may decide to hire a manager whose job is to know who is working on what, and to then allocate responsibilities among team members. 3

In related work, Fellner et al. (2011) study a public goods game with heterogeneous productivity types, high and low, and vary whether a) subjects are aware of the heterogeneity and b) whether subjects can link contributions to types. The authors also find that subjects exploit information asymmetries. When subjects are aware of the heterogeneity, but cannot link contributions to types, high types contribute less. 4 Corroborating evidence for this preference is provided by Ertan et al. (2009), who observed groups as they voted on how to implement peer punishment in a public goods game. By and large, groups elected to punish only low contributors, while no group ever allowed the punishment of high contributors.

2

30 31 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

In these examples, the advantage of a central authority is that it has better information about endowments than any one individual. The local government has relevant tax information tied to income or property value, while the manager of the team observes how employees allocate their time towards various projects. Accordingly, we designed our central authority along these lines. Specifically, it has perfect information and imposes the socially-optimal outcome, but does so at a moderately high fixed cost. Therefore, groups that are able to effectively deploy peer punishment have an incentive to eschew the central authority. The question then is whether groups can effectively use peer punishment when subjects have private information about endowments. In this paper, we show there is a link between incomplete information and a preference for central authority when there is income inequality within groups. Income inequality is implemented by exogenously varying subject endowments.5 We develop three, linear public good experiments: one in which each group member receives an equal endowment, and two in which group members receive unequal endowments: Low (10), Medium (20) and High (30). Across the unequal and heterogeneous endowment treatments we alter the information about endowments available to group members. In our Observed treatment, group members always observe group member endowments alongside their contributions after each period. In our Unobserved treatment, group members only observe the contributions of group members, and receive no information concerning their endowment. In all treatments, a subject always knows their own endowment, as well as the distribution of endowments across group members.6 After gaining experience with both peer punishment and central authority, groups vote for an enforcement institution – central authority or peer punishment – for the remainder of their session. In contrast to Nicklisch et al. (2016), our alternative institution to peer punishment is a central authority that strictly enforces the social optimum. As such, incomplete information primarily impacts the effectiveness of the peer punishment mechanism. In this sense, we are not investigating behavior across institutions; rather, we create an alternative which ought to be equally effective across all treatments and investigate how the relative effectiveness of peer punishment alters institutional preference across treatments. Therefore our design complements Nicklisch et al. (2016) and motivates incomplete information as another plausible mechanism through which central authority institutions have come to dominate modern society. 5

In this paper we use the terms “income” and “endowment” interchangeably. Note that complete versus incomplete information here refers to whether subjects observed each other’s incomes. In all our treatments subjects perfectly observe each other’s contributions. 6

3

30 31 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31

Our main finding is that groups with income inequality and incomplete information tend to choose central authority over peer punishment, largely due to the votes from subjects with lower incomes. At the same time, heterogeneous groups with complete information tend to prefer peer punishment at the same rates as homogeneous groups. Therefore, as suggested in Reuben and Riedl (2013), endowment heterogeneity alone does not drive the preference for central authority. Instead, this preference appears to stem directly from the ineffectiveness of peer punishment to improve overall group welfare, particularly for subjects with lower incomes. Consistent with other studies, our results suggest peer punishment was ineffective because it could not be targeted at freeriders. Subjects with high incomes earned similar payoffs across information treatments. By contrast, subjects with lower incomes earned significantly less under incomplete information, and were often met with sanctions, even when they contributed their full endowments. The costs of incomplete information thus fell mostly on subjects with lower incomes, who were discouraged to cooperate, while subjects with higher incomes were encouraged to freeride. When endowments were observed, targeting improved, and both the disincentive to freeride and incentive to cooperate were restored. While the direct contribution of this paper is to the experimental literature on social dilemmas, our results extend to real-world scenarios that involve consensus and enforcement. Our study complements recent work on income inequality in social dilemmas, investigating the idea that income inequality combined with information asymmetries can undermine selfgovernance. For instance, research on economic development suggests that societies with higher levels of income inequality under-invest in public goods like education (Easterly, 2007). Similarly, studies on micro-finance in developing countries find that group lending is more successful when groups are homogeneous; since micro-finance loans rely on peer monitoring and enforcement to ensure repayment, homogeneous groups are better at creating consensus and enforcing it (Cassar et al., 2007). In general, recent experimental evidence suggests that income inequality limits the ability of groups to reach consensus (e.g. about what constitutes fairness) and therefore limits their capacity for self-governance in settings with asymmetric information (Kingsley, 2016). Our results suggest that groups in these environments may seek out costly central authority regimes that can mitigate the information asymmetries and promote cooperation. The rest of our paper proceeds as follows. Section 2 presents our experiment design and hypotheses. Section 3 presents and discusses our results. Finally, Section 4 concludes and proposes future directions for this research. 4

32

33 1 2 3 4 5 6 7 8 9

2. Experimental design and procedures The experiment consisted of three treatments: a control in which endowments were equal and thus observed (Equal ), a treatment in which endowments were unequal and unobserved (Unobserved ), and finally a treatment in which endowments were unequal and observed (Observed ). Each treatment included eighteen periods broken into six phases of three periods each and was coded in z-Tree (Fischbacher, 2007). Groups of five subjects were randomly formed at the beginning and maintained for all eighteen periods. Instructions for each phase were distributed at the beginning of each phase and required each participant to correctly answer a set of comprehension questions before the experiment would continue.7 The sections below describe each phase of the experiment. A summary illustration of procedures within a treatment is provided here in Figure 1. Exogenous institutions VCM

PP/CA

PP/CA

(Periods 1-3)

(Periods 4-6)

(Periods 7-9)

Phase 1

Phase 2

Phase 3

Endogenous institutions

Vote

PP/CA Phase 4

(Periods 10-12)

PP/CA

Vote

Phase 5

(Periods 13-15)

Vote

PP/CA Phase 6

(Periods 16-18)

Figure 1: Summary of our treatment phases. Each group participated in six phases. The first phase (periods 1-3) was a VCM (i.e. no punishment). In phases 2 (periods 4-6) and 3 (periods 7-9) either peer punishment or central authority was exogenously imposed, with the order of imposition reversed across sessions. Groups could then vote on three occasions for an institution to be implemented in phases 4 (periods 10-12), 5 (periods 13-15), and 6 (periods 16-18).

10 11 12 13 14 15

2.1. Voluntary contribution baseline Phase 1 (periods 1 - 3) in all treatments introduced the standard voluntary contribution mechanism (VCM), wherein subjects decided how much to contribute to a group account. Each subject was given a fixed endowment of experimental currency (herein referred to as experimental dollars, EDs) which could be allocated between a private account and a group account. Payoffs to subject i were πi = (ei − xi ) + α

16 17

n X

xj

(1)

j=1

where xi is the subject’s contribution to the group account, ei is the subject’s endowment, Pn α = 0.4 is the marginal per capita return (MPCR) from the public good, and j=1 xj 7

Complete experimental instructions are available in our online appendix.

5

18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

represents the sum of contributions to the group account from all group members. With n players, n1 < α < 1, and a known last period, there is a unique, symmetric Nash equilibrium where everybody freerides and contributes nothing to the public good. Similarly, there is a symmetric social optimum where subjects contribute their entire endowment to the public good. In the baseline treatment each participant received the same, fixed, endowment of 20 EDs. We will refer to this treatment as Equal. In the other two treatments the endowments were heterogeneous. In both heterogeneous treatments the distribution of group endowments was identical and known. Each group was composed of two High endowment members who received 30 EDs, one Middle endowment member who received 20 EDs, and two Low endowment members who received 10 EDs. Endowments in the heterogeneous treatments were randomly assigned and maintained for the entire experiment (i.e. once a High type always a High type). Note that the aggregate level of endowments, or the resources available P for contributing towards the public good, was the same across all three treatments ( ni=1 ei = 100 EDs).8 In all treatments, subjects were shown the following information after each group member had made their contribution decision: aggregate contribution to the group account; the individual contributions of their group members by random ID; their individual period earnings; their total earnings (equal to the sum of their individual period earnings); and a history of outcomes in previous periods.9 Additionally in the heterogeneous endowment treatments, group members either observed or did not observe the endowment of each group member after each period. For example, in the Observed treatment, group members not only observed the contribution decisions but they also observed that group member’s endowment. In contrast, in the Unobserved treatment, group members would only observe the contribution decisions and would not observe the group member’s endowment. The experimental literature on public goods has established that contributions tend to lie between the Nash and social outcomes, and decline with repetition (Davis and Holt, 1993; Ledyard, 1995; Chaudhuri, 2011). In the absence of any institutional constraints on behavior, results from linear public good experiments suggests that endowment heterogeneity 8

It is important to hold the aggregate level of endowments constant in order to compare earnings across conditions. With this in mind, and as discussed below, our use of majority voting, adding a single middle endowment type was the most straightforward way to introduce endowment heterogeneity without tilting the vote towards either endowment type (for example, having three low types or three high types). 9 The random ID and the order of presentation of the contributions of one’s group members was randomized each period to avoid reputation effects.

6

28 29 30 1

has no effect on average contributions or earnings (Buckley and Croson, 2006; Reuben and Riedl, 2013). Given the lack of a behavioral response to endowment heterogeneity, the availability of endowment information is not expected to effect behavior. This leads to our first hypothesis:

3

Hypothesis 1. In the VCM periods, average earnings will not be statistically different across the Equal, Observed, or Unobserved treatments.

4

2.2. Exogenously imposed institutions

2

10

During phase 2 (periods 4 - 6) and phase 3 (periods 7 - 9), each group was introduced to two types of institutions, peer punishment and central authority, explained in detail below. To account for order effects, the order of imposition was reversed across the six experimental sessions conducted for each treatment. Therefore, approximately half of all groups in each treatment had the peer punishment mechanism imposed in phase 2 followed by the central authority mechanism in phase 3, while the order was reversed for the other half.10

11

2.2.1. Peer punishment mechanism

5 6 7 8 9

12

Peer punishment alters i’s payoffs as follows: πi = (ei − xi ) + α

13 14 15 16 17 18 19 20

n X j=1

xj − c

n−1 X j6=i

Pij − r

n−1 X

Pji

(2)

j6=i

where Pij represents the number of reduction points that i imposes on other group members j at a cost of c = 1, and Pji represents the number of reduction points that other group members j impose upon i at a cost of r = 4.11 In order to avoid excessive losses subjects were allowed to impose up to 10 reduction points per period and, similar to Gächter et al. (2008) and Reuben and Riedl (2013), losses on any given period were bounded at zero unless the subject imposed punishment.12 The costs associated with imposing “Reduction Points” were referred to as “Administrative Costs” and costs associated with receiving “Reduction Points” were referred to as “Reduction Costs.” 10

Specifically, 11, 11, and 12 groups in Equal, Observed, and Unobserved participated in peer punishment followed by central authority while 12 groups in all treatments participated in the central authority followed by peer punishment. We find no evidence of order effects. These results can be found in our online appendix. 11 Nikiforakis and Normann (2008) and Egas and Riedl (2008) note that the punishment technology must be at least 1:3 (pay one ED to reduce a group member’s payoff by 3 EDs) to be effective. 12 Accordingly, payoffs under the peer punishment mechanism are in fact πi = max[0, (ei − xi ) + Pn Pn−1 Pn−1 α j=1 xj − r j6=i Pji ] − c j6=i Pij .

7

21 22 23 1 2 3 4 5 6 7 8 9

10 11

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 1

Since punishment is costly to deploy, group welfare only improves when contributions rise sufficiently to offset these costs. Research suggests that peer punishment tends to be effective when contribution norms are unambiguous and thus deviations are easily observed and can be punished (Gurerk et al., 2006; Nikiforakis et al., 2012; Reuben and Riedl, 2013; Neitzel and Saaksvuori, 2013; Kingsley, 2016). In public good experiments several plausible contribution norms may emerge. Reuben and Riedl (2013) suggest that efficiency, payoff equality, and contribution equality are all plausibly appealing. Further, investigating the effect of endowment heterogeneity on peer punishment with complete information, similar to our Observed treatment, Reuben and Riedl (2013) suggest that groups successfully deploy punishment to increase contributions and induce cooperation.13 This suggests that endowment heterogeneity alone will not substantially hinder the effectiveness of peer punishment: Hypothesis 2. Average earnings within the peer punishment mechanism will not be statistically different across the Equal and Observed treatments. The importance of contribution norms is that punishment can be effectively targeted at those who deviate from the norms. As mentioned above, Grechenig et al. (2010) and Ambrus and Greiner (2012), investigate the impact of imperfect information on peer punishment. When contributions are observed with error, peer punishment is found to lower group welfare below their no-punishment baseline levels. Results suggest that when freeriders were punished they did not increase their subsequent contributions. This suggests that the received punishment was either too mild to alter incentives or that freeriders expected imperfect information to shield them from future punishment. Further, when cooperators were punished they tended to decrease their subsequent contributions. In this sense, imperfect information mitigates the costs of freeriding as well as the benefits of cooperation under peer punishment. Following this logic, incomplete information is expected to have a similar effect on peer punishment. In Unobserved there is an opportunity for group members with higher endowments to conceal their status by lowering their contribution toward levels observed by group members with lower endowments. Thus, there is an opportunity for freeriding high members to avoid punishment, and there is also an opportunity for cooperating low members 13

In linear public good experiments with boundary solutions and homogeneous endowments, the contribution norms of efficiency, payoff equality, and contribution equality all coincide. Similarly, even with endowment heterogeneity norms of efficiency and payoff equality continue to coincide. Nonlinear public good experiments can be used so that norms of payoff equality and contribution equality can be independently efficient to differentiate between these competing norms (Kingsley, 2016).

8

2 3

to be punished. If behavior under incomplete information is similar to that observed under imperfect information we expect peer punishment to be less effective:

5

Hypothesis 3. Average earnings within the peer punishment mechanism will be significantly higher in the Equal and Observed treatments relative to the Unobserved treatment.

6

2.3. Central authority

4

7

Finally, central authority alters i’s payoffs as follows: πi = (ei − xi ) + α

8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

24 25

n X j=1

xj − c − ps (ei − xi )

(3)

where c = 6 is the fixed cost of implementing the central authority for each group member, s = 1.2 is the level of the sanction, and p = 1 is the probability of being monitored (we set p = 1 without loss of generality and for simplicity).14 The sanction is proportional to the deviation from the social optimum contribution (e − xi ). Since ps > 1 − α, the expected cost of freeriding is strictly greater than the benefit and the institution is deterrent.15 The intuition is that a deterrent institution will alter incentives such that it is in an individual’s self-interest to contribute to the public good (Becker, 1968; Stigler, 1970; Polinsky and Shavell, 1979; Ehrlich, 1996). Within the public goods literature, deterrent central authority mechanisms have been shown to increase contributions and, depending on the fixed costs, to increase group welfare (Kamei et al., 2015; Markussen et al., 2014). For consistency across institutions the fixed costs associated with the central authority were referred to as “Administrative Costs” and any sanctions imposed by the central authority were referred to as “Reduction Costs.” Since aggregate group endowments were equivalent across treatments and deterrence alters each group members incentives to contribute their entire endowment we have the following hypothesis: Hypothesis 4. Average earnings within the central authority will not be statistically different across the Equal, Observed, or Unobserved treatments. 14

These parameters, as discussed in Section 2.4, were chosen to mirror the relevant institutional choice literature. 15 There are various ways researchers have designed the deterrence of central authority mechanisms. Galbiati and Vertova (2008) consider minimum contribution obligations, Tyran and Feld (2006) considered fixed sanctions that were not proportional to one’s contribution, while Kamijo et al. (2014) and Andreoni and Gee (2012) consider mechanisms that only punish the worst offender (lowest contributor). Regardless of the methodology, results suggest that deterrent central authority mechanisms effectively increase contributions.

9

26 27 28 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

2.4. Institutional choice After these institutions were exogenously imposed each group had three opportunities to choose their institution before phases 4, 5, and 6. Voting was by majority and each subject had to vote to operate within either the peer punishment or central authority mechanism.16 Research suggests that groups will self-impose deterrent central authority institutions when it benefits them, but that they are sensitive to the fixed costs associated with the central authority (Kamei et al., 2015; Markussen et al., 2014). In Kamei et al. (2015) subjects chose between operating within a peer punishment or a central authority mechanism. The central authority monitored all subjects with certainty but the level of the sanction (proportional to the deviation from the socially optimum contribution) was determined through a voting procedure. The level of the sanction could be non-deterrent (s ∈ {0, 0.4}) or deterrent (s ∈ {0.8, 1.2}). Across treatments, the fixed cost of the institution could be low (c = 0) or high (c = 5). When the central authority was costless (costly) a large majority of groups favored central authority (peer punishment). In Markussen et al. (2014), groups could chose to operate within a central authority, a peer punishment mechanism, or a VCM. Across four treatments they manipulated the parameters of the central authority that sanctioned, with certainty, the deviation from contributing one’s entire endowment. The sanction level was either deterrent (s = 0.8) or non-deterrent (s = 0.4) and the cost was either low (c = 2) or high (c = 8). Results suggest that only the deterrent and low cost central authority was preferred to the VCM and to the peer punishment mechanism. In both papers, groups tended to choose the institution that increased their earnings. Our central authority was designed to mirror this literature. We designed our central authority to be deterrent and to enforce the social optimum, but it does so at a relatively high fixed cost. The central authority presented here solves the problems associated with misguided or ineffectively targeted peer punishment: freeriders will be punished such that their self-interest is to contribute fully towards the public good and cooperators will have no possibility of punishment. Therefore, our central authority offers groups an effective alternative to peer punishment. But it comes at a relatively high cost, so groups will be better off if they can effectively deploy peer punishment. The benefit of simplifying the central authority is that variation in institutional preference should be driven entirely by the relative effectiveness of peer punishment across treatments. 16

We used majority voting for its ease of explanation and to follow Markussen et al. (2014) and Kamei et al. (2015) since their studies are closest to ours. An alternative to majority voting is to employ consensus voting (Kosfeld et al., 2009; Sutter et al., 2010).

10

30 31 32 33

1 2

Following the hypotheses stated above, if the peer punishment mechanism is proven to be less effective in Unobserved, relative to Observed, while the central authority mechanism is equally effective, the expectation is that more groups will choose to self-impose the central authority in Unobserved relative to Observed : Hypothesis 5. The proportion of groups choosing to operate within the central authority mechanism will be greater in Unobserved relative to Equal and Observed treatments.

7

Finally, it is important to highlight the trade-off in our design with regards to the number of periods in each phase and the number of phases. Shorter phases allowed us to give subjects experience with both institutions and multiple voting opportunities. However, this limits our analysis to the short-run dynamics we can observe in three periods. In the discussion we point out how future work can build on this design to examine long-run patterns.

8

3. Results

3 4 5 6

9 10 11 12 13 14 15 16 17 18

1 2 3 4 5 6

Data was collected in April, May, November, and December of 2016 at the University of Massachusetts Amherst Cleve E. Willis Experimental Economics Laboratory. A total of eighteen sessions were conducted (350 participants in 70 groups). The average session lasted approximately 75 minutes. Our analysis is restricted to behavior in the exogenous phases (Phases 1 - 3 in Figure 1). We organize our results into two sections. First, we take a look at payoffs and voting behavior across treatments; this section will serve to test our main hypotheses. We will show that a clear picture comes into focus: when endowments cannot be observed, groups eschew peer punishment for central authority. The second section of our results will try to explain how this shift in preferences for enforcement institutions may have come about.17 3.1. Overall results In Table 1 we report the average group payoffs across our three treatments in each exogenously imposed institution. We compare our results across Equal /Observed /Unobserved for each enforcement institution (VCM, PP and CA). To compare average payoffs we use Kruskal-Wallis tests.18 Average group payoffs are statistically equivalent across all treatments in the VCM (χ22 = 3.31, p = 0.191), consistent with our first hypothesis. In the 17

The data and code for our analysis can be found at https://github.com/lrdegeest/ InstitutionalChoice. 18 Each Kruskal-Wallis and Wilcoxon Rank-Sum test presented are conducted using the group average as the unit of observation. We have 23, 23, and 24 group level observations in Equal, Observed, and Unobserved respectively.

11

7 8 9 10 11 12

absence of any institutional constraints on behavior, endowment heterogeneity has no significant effect on behavior. Turning to the central authority, column 3 in Table 1, we see a weakly significant difference across group payoffs (χ22 = 5.01, p = 0.082). However, when only the last period of each exogenous phase is compared this difference falls to insignificance (χ22 = 3.46, p = 0.177). This is expected given the deterrence of the central authority, and consistent with our fourth hypothesis. Table 1: Average Payoffs.

N Equal Observed Unobserved

23

VCM

32.92 (2.98) 23 31.14 (4.03) 24 30.39 (4.92)

Peer Punishment

Central Authority

27.52 (6.01) 29.52 (6.25) 24.10 (6.20)

31.32 (2.66) 31.56 (2.03) 29.93 (2.72)

Average group earnings in each exogenously imposed institution. Standard deviations are presented in parentheses.

13 14 15 16 17 18 19 20 21 22 1 2 3 4

We now turn to our second and third hypotheses. Comparing the payoffs in the peer punishment mechanism, we see a significant difference across the three treatments for all periods (χ22 = 7.25, p = 0.027), and for just the last period (χ22 = 17.22, p = 0.01). As expected, there is no significant difference in payoffs under peer punishment when endowments are equal or when endowments are heterogeneous and observed (Equal vs. Observed z = 1.02, p = 0.31). However, payoffs are significantly lower in Unobserved relative to the other two treatments (Equal vs. Unobserved z = 1.74, p = 0.08; Observed vs. Unobserved z = 2.60, p = 0.01). Our results suggest that endowment heterogeneity, in linear public good experiments with complete information, does not significantly alter the successful use of peer punishment. However, under incomplete information, peer punishment is less successful.19 The relative success of peer punishment has a direct bearing on group preferences for institutions. Each group, in each treatment, voted prior to phase 4, phase 5, and phase 6. Across Equal, Observed, and Unobserved, we have 69, 69, and 72 group level voting observations, respectively. To quantify the relationship between treatments and preference for 19

These non-parametric results are backed up by linear random effects regressions on final and initial payoffs. Results are available in our online appendix.

12

5 6 7 8 9 10 11 12

central authority, χ2 tests are used to determine whether a group’s treatment is independent of their voting preference. As shown in Panel A of Figure 2, there is a clear preference for the central authority when endowments are heterogeneous and unobserved (Equal vs. Unobserved : χ2 = 19.63, p = 0.01; Observed vs. Unobserved, χ2 = 24.68, p = 0.01). Overall, groups in Unobserved self-imposed the central authority on 46 out of 69 (67%) voting opportunities. By contrast, the proportion of groups imposing central authority in Equal and Observed was 29% (20 out of 69) and 25% (18 out of 72), respectively. Indeed, there is no difference in voting behavior across Equal and Observed (χ2 = 0.28, p = 0.59).20 A

100

B

Unobs.

Percent of total

80

Obs.

60

40

20

Equal

0

20

40

60

80

0

100

Unobs.Obs.

Unobs.Obs.

Unobs.Obs.

Low

Middle

High

Percent of total

PP

CA

Figure 2: Proportion of votes for the central authority (CA) and peer punishment (PP). Panel A displays aggregate voting across treatments. Panel B displays voting by endowments in the heterogeneous endowment treatments.

13 14 15 16 17 18

Importantly, the preference for central authority is not symmetric across endowments. As shown in Panel B of Figure 2, the difference in voting between Observed and Unobserved is driven largely by the Low and Middle types. Significantly more Low and Middle types voted for the central authority in Unobserved vs. Observed (Low: χ2 = 44.5, p = 0.01; Middle: χ2 = 23.67, p = 0.01). However, there is no significant shift in voting among High types (χ2 = 0.79, p = 0.37). 20

We do not observe large changes in how groups vote across phases. The proportion of groups choosing central authority in phase 4, phase 5, and phase 6 was 0.30, 0.26, and 0.30 in Equal ; 0.29, 0.25, and 0.21 in Observed ; and 0.70, 0.70, and 0.61 in Unobserved.

13

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 1 2 3 4 5

The institutional choice literature suggests that subjects vote for the institution under which they earn more (Markussen et al., 2014; Kamei et al., 2015). To compliment the above analysis we present a probit model to investigate the impact of the earnings difference across institutions on one’s vote for the central authority in Table 2. The difference in earnings is calculated as the difference between the average that the subject earned during the three periods of the central authority and the average earned during the three periods of peer punishment. We also include an Observed indicator to account for a treatment effect. The first set of models – (1) to (4) – focus on the first voting opportunity, prior to phase 4. The second set of models – (5) to (8) – look at all three voting phases. Models (1) and (5) are aggregated across endowments; the remaining models are broken down by endowment. In each specification, the earnings differential has a significant effect on voting choice: subjects who earned more in the central authority institution tended to vote for it. Moreover, institutional choices do not appear to change in subsequent voting opportunities, suggesting groups tended to stick with the institution they chose first. The only exception is the significant effect for Phase 5 among Middle; however, the significance fades in the next voting phase. Turning to the treatment effect, we see that being in Observed reduced the probability of self-imposing the central authority. While coefficients are negative across the board, in the First Vote models, the effect is only significant for All types and Middle types. However, in the All Votes models, the effect is significant in each specification, except among High types.

14

Table 2: Earnings differential and voting for central authority. First Vote

Earning Difference Observed

All Votes

(1) All

(2) Low

(3) Middle

(4) High

(5) All

(6) Low

(7) Middle

(8) High

0.080∗∗∗ (0.01) -0.341∗ (0.18)

0.061∗∗∗ (0.02) -0.517 (0.32)

0.095∗∗ (0.04) -1.044∗∗ (0.49)

0.099∗∗∗ (0.02) -0.113 (0.29)

-0.697∗∗∗ (0.20)

-0.334 (0.38)

-0.770 (0.73)

-1.071∗∗∗ (0.38)

0.063∗∗∗ (0.01) -0.510∗∗∗ (0.18) -0.071 (0.10) -0.090 (0.12) -0.404∗∗ (0.18)

0.057∗∗∗ (0.02) -0.612∗∗ (0.28) -0.151 (0.15) -0.184 (0.17) -0.163 (0.32)

0.061∗∗ (0.03) -1.075∗∗∗ (0.35) 0.366∗∗ (0.18) 0.142 (0.26) 0.229 (0.51)

0.067∗∗∗ (0.01) -0.245 (0.29) -0.194 (0.13) -0.118 (0.14) -0.806∗∗ (0.32)

Yes 235 0.212

Yes 94 0.210

Yes 47 0.326

Yes 94 0.256

Yes 705 0.171

Yes 282 0.200

Yes 141 0.218

Yes 282 0.162

Phase 5 Vote Phase 6 Vote Constant Demographics Controls N Pseudo R-squared

Demographic controls: Gender (Male/Female), number of undergraduate semesters, and number of undergraduate economics classes. Standard errors clustered at the group level. ∗

6 7 8 9 10 11 12 13 14 15 16 17 18

p < 0.10,

∗∗

p < 0.05,

∗∗∗

p < 0.01

3.2. The role of information in peer punishment Our results so far speak to a straightforward outcome: under incomplete information, central authority is preferred to peer punishment. However, it is not immediately evident why this preference would be driven by subjects with lower endowments. Here we explore the role of information in peer punishment and how it alters institutional choice. Since the central authority imposes a constant, fixed cost, subjects could theoretically improve their lot in peer punishment by establishing and enforcing a contribution norm. Our focus here is comparing Observed with Unobserved in peer punishment. As noted above, the role of information in the central authority regime here is minimal.21 Figure 3 shows the average contributions, the average contributions as a percentage of endowment, and the average payoffs for each endowment in Observed and Unobserved under peer punishment. Aggregating across all endowments we can see that both average contributions and average payoffs are significantly greater with complete information. Comparing 21

Results from our central authority phases, in which the central authority enforced the social optimum, can be found in our online appendix.

15

2 3 4 5 6 7 8

Observed vs. Unobserved, we observe average contributions of 16.1 vs. 11.9 (z = 3.62; p < 0.01) and average payoffs of 29.5 vs. 24.1 (z = 2.60; p < .01). However, the impact of information is not symmetric across endowments. The impact of observing endowments within the peer punishment mechanism is immediately clear: it increases the contributions of High members and substantially reduces the level of payoff inequality across endowments. Comparing Observed vs. Unobserved we observe average contributions of 8.7 vs. 7.3 among Low members (z = 1.63; p = 0.103), 15.8 vs. 13.5 among Middle members (z = 1.14; p = 0.256), and 23.7 vs. 15.6 among High members (z = 3.66; p < 0.01).22 30

A

100

Contribution (% of endowment)

25

Contribution

20

15

10

35

C

80 30

60 25

40

20 20

5

0

B

Payoff

1

Unobserved

Observed

0

Unobserved

Low

Middle

15

Observed

Unobserved

Observed

High

Figure 3: Average contributions and payoffs across endowments and Observed/Unobserved, peer punishment and exogenous groups. Error bars show the standard error of the mean.

9 10 11 12 13 14 15 16 17 18 1

Importantly, both the Low and Middle types are significantly better off within peer punishment when endowments are observed while High members are not significantly affected. Specifically, comparing Observed vs. Unobserved average payoffs are 29.6 vs. 18.3 among Low members (z = 4.01; p < 0.01), 30.2 vs. 23.6 among Middle members (z = 2.83; p < 0.01), and 29.1 vs. 30.1 among High members (z = 0.543; p = 0.587). To probe further, we looked at how much punishment was deployed and how it was allocated. Overall, a similar amount of punishment was deployed across treatments (Observed vs. Unobserved : 1.34 vs. 1.59; z = 0.405; p = 0.686). However, interesting patterns emerge when we look at sanctions supplied and received by each endowment. Figure 4 breaks down the supply and receipt of sanctions across endowments and information settings. Panel A displays the average amount of punishment sent (sanctions) while Panel B displays the average amount 22

Analysis of the proportional contributions is statistically equivalent (Overall: 0.82 vs. 0.64, p < .01; Low: 0.87 vs. 0.71, p = .10; Middle: 0.79 vs. 0.68, p = 0.26; High: 0.79 vs. 0.52, p < .01).

16

of punishment received (sanctions × 4).

Average sanction sent

12

A

12

Average cost of sanctions received (sanction x 4)

2

8

4

0

Unobserved

Low

8

4

0

Observed

Middle

B

Unobserved

Observed

High

Figure 4: Sanctions supplied (Panel A) and the cost of sanctions received (Panel B) by endowment and Observed/Unobserved. Error bars show the standard error of the mean.

3 4 5 6 7 8 9 10 11 1 2 3 4 5

Information seemed to have little impact on how sanctions were supplied. Panel A suggests that across Observed and Unobserved there is no statistical difference in the average amount of punishment sent from Low members (1.45 vs. 1.69; z = 0.011; p = 0.99), from Middle members (1.14 vs. 1.78; z = 1.364; p = 0.17), or from High members (1.37 vs. 1.39; z = 0.053; p = 0.96). On the other hand, as suggested in Panel B, information about endowments seemed to impact how sanctions were allocated, specifically reducing the amount of punishment received by Low members (Observed vs. Unobserved : 2.50 vs. 6.70; z = 2.713; p < 0.01). With no significant differences observed across Middle members (5.28 vs. 4.99; z = 0.194; p = 0.85) or High members (8.42 vs. 6.70; z = 1.066; p = 0.29). While it appears that subjects were better able to target sanctions at freeriders under complete information, a regression analysis suggests that the impact of information is more nuanced. Table 3 estimates subject payoffs controlling for a subject’s contribution, the contributions of others in one’s group, a time trend (Period), an Observed indicator, and an interaction between one’s contribution and Observed.

17

Table 3: Estimating the effect of information on payoffs in peer punishment.

Low

Contribute

Middle

(1)

(2)

(3)

(4)

(5)

(6)

0.885∗∗∗ (0.25) 0.296∗∗∗ (0.04) 4.360∗∗∗ (1.49) -0.519 (0.48)

-0.061 (0.24) 0.317∗∗∗ (0.07) 0.026 (2.05) -0.085 (0.71)

-0.012 (0.10) 0.243∗∗∗ (0.05) -3.998∗∗ (1.87) 1.780∗∗ (0.85)

14.150∗∗∗ (3.71)

-0.423 (0.30) 0.328∗∗∗ (0.07) -10.430 (7.01) -0.305 (0.71) 0.697∗ (0.37) 19.257∗∗∗ (4.81)

14.930∗∗∗ (3.89)

-0.292∗∗∗ (0.09) 0.229∗∗∗ (0.04) -18.468∗∗∗ (5.52) 1.136 (0.83) 0.722∗∗∗ (0.20) 20.378∗∗∗ (3.68)

Yes 141 0.371

Yes 141 0.422

Yes 282 0.121

Yes 282 0.211

Constant

-1.649 (2.90)

0.715∗∗ (0.34) 0.294∗∗∗ (0.04) 0.418 (4.20) -0.597 (0.46) 0.486 (0.42) -0.096 (3.44)

Demographics Controls N Adjusted R-squared

Yes 282 0.655

Yes 282 0.659

Others contribution Observed Period

High

Observed X Contribute

Demographic controls: Gender (Male/Female), number of undergraduate semesters, and number of undergraduate economics classes. Standard errors clustered at the group level. ∗

6 7 8 9 10 11 12 13 14 15

p < 0.10,

∗∗

p < 0.05,

∗∗∗

p < 0.01

First, we consider the results presented in columns (1), (3), and (5). This specification holds the marginal effect of one’s own contribution constant across treatments for each endowment and allows us to interpret the coefficient on Observed as the average marginal effect of the treatment. We will relax this assumption next but it is worth noting two effects across endowments. First, all else equal, a Low member can expect to earn 4.36 more, a Middle member can expect to earn a statistically equivalent amount, and a High member can expect to earn 4.00 less in Observed. Second, the marginal effect of increasing one’s own contribution is only positive among Low members, while it is negative and insignificant among Middle and High members.23 As expected, an increase in the contributions from the other members of one’s group increases one’s payoff across all endowments. 23

Controlling for one’s contribution is a straightforward way to determine whether peer punishment was

18

16 17 18 19 20 21 22 1 2 3 4 5 6 7 8 9 10 11 12

Columns (2), (4), and (6) include an interaction term between one’s contribution and Observed. Note, that the positive effect of the contributions of others remains. The coefficient on Contribute now represents the marginal effect of increasing one’s contribution within Unobserved, and the coefficient on the interaction term represents the change in the marginal effect of increasing one’s contribution in Observed. Combining terms and estimating the average marginal effect of one’s contribution in Observed, we observe that it is positive and significant for Low and High members: Low (1.20 (0.27), p < 0.01), Middle (0.27 (0.29), p = 0.34), and High (0.43 (0.19), p = 0.03).24 The results for Observed are consistent with our expectations: low contributions are punished and group members, at least Low and High types, are significantly better off contributing more. By contrast, peer punishment in Unobserved produced less consistent results, as we see that increasing one’s contribution has a positive, insignificant, and negative effect on one’s payoffs among Low, Middle, and High members respectively.25 To provide greater context for the effect of information within the peer punishment regime, Figure 5 shows the predicted difference in earnings (Observed minus Unobserved ) by one’s contribution using the models presented in (2), (4) and (6) of Table 3. There are two significant results displayed. First, cooperating Low types (those contributing their entire endowment) earn significantly less in Unobserved.26 Second, freeriding High types, and more generally those contributing less than 20, earn significantly more in Unobserved.

effectively deployed. In the absence of peer punishment, increasing one’s contribution must necessarily reduce one’s earnings in a public goods game. However, if peer punishment is targeted at low contributions effectively than increasing one’s contribution should increase one’s earnings. 24 When marginal effects are reported within the text, standard errors will be reported in parentheses. 25 At first pass, the positive effect of Contribute among Low types in Unobserved appears inconsistent. Without endowment information, why would only Low types be rewarded by contributing more? Upon further analysis it was determined that contributions up to ten were treated similarly across all endowments in Unobserved. In our online appendix we present estimates of earnings and punishment received in Unobserved for contributions less than or equal to ten across endowments. The main effect (Contribute) is significant and positive for earnings and negative for punishment received, while the interactions with Middle and High are insignificant. 26 A similar, but less pronounced, effect is found among Middle types. The predicted difference in earnings for Middle types who contribute their entire endowment is significant at the 10% level (p = 0.07).

19

Middle

High 10

0

0

0

-10

-20

-30

Predicted difference in earnings

10

Predicted difference in earnings

Predicted difference in earnings

Low 10

-10

-20

-30 0

5

Contribution

10

-10

-20

-30 0

10

Contribution

20

0

10

20

30

Contribution

Figure 5: Average marginal effects for Table 3 by one’s contribution. The vertical axis shows the predicted difference in punishment in Observed relative to Unobserved. The shaded region is the 95 percent confidence interval.

13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 1 2

To summarize, these results suggest that the impact of information depends largely on one’s income. Recall that peer punishment is ineffective, even harmful, when it cannot be targeted. There are two sources of this harm. First, when punishment cannot be targeted at freeriders, it reduces the costs of freeriding and limits the incentives for those punished to increase their contributions. Second, when punishment is aimed at cooperators, it reduces the benefits of cooperation and limits their incentives to continue contributing. With incomplete information, this intuition takes a very specific form: cooperative, lower endowment members (those contributing their entire endowments) are significantly better off in Observed. Put another way, cooperative, lower endowment members earn significantly less when information is incomplete, indicating that they are targeted despite cooperating. By contrast, this relationship is reversed for High members. Non-cooperative High members earn significantly less in Observed, or alternatively, significantly more in Unobserved. This implies that High members are able to avoid punishment while freeriding when information is incomplete. The above regression analysis on payoffs across endowments suggests that the cost of incomplete information falls largely on lower income group members. Despite contributing similar, and largely cooperative, amounts in both treatments, Low and Middle types earn significantly less in Unobserved relative to their Observed counterparts. We have also observed that while the overall deployment of punishment is similar across treatments, the allocation of punishment is significantly altered across endowments. This suggests that 20

3 4 5 6 7 8

punishment is ineffectively deployed when information is incomplete. To investigate this directly we present a similar regression to the one above to understand the determinants of punishment received. Like the previous regression, Table 4 estimates punishment received controlling for a subject’s contribution, the contributions of others in one’s group, a time trend (Period), an Observed indicator, and an interaction between one’s contribution and Observed. Table 4: Estimating the effect of information on punishment received in peer punishment.

Low

Contribute Others contribution Observed Period

Demographics Controls N Adjusted R-squared

High

(1)

(2)

(3)

(4)

(5)

(6)

-1.730∗∗∗ (0.27) 0.129∗∗∗ (0.03) -4.521∗∗∗ (1.31) 0.749 (0.45)

-0.602∗∗ (0.23) 0.116∗ (0.06) 0.046 (1.74) 0.521 (0.69)

3.173 (3.40)

-0.209 (0.25) 0.104∗ (0.06) 11.411∗ (6.51) 0.760 (0.69) -0.758∗∗ (0.36) -2.377 (4.47)

-0.706∗∗∗ (0.11) 0.224∗∗∗ (0.06) 4.381∗∗ (1.72) -1.927∗∗ (0.82)

10.500∗∗∗ (2.82)

-1.627∗∗∗ (0.37) 0.130∗∗∗ (0.03) -2.135 (4.41) 0.797∗ (0.43) -0.294 (0.44) 9.559∗∗∗ (3.53)

12.718∗∗∗ (3.59)

-0.406∗∗∗ (0.09) 0.239∗∗∗ (0.05) 19.931∗∗∗ (6.10) -1.235 (0.77) -0.776∗∗∗ (0.23) 6.865∗ (3.54)

Yes 282 0.465

Yes 282 0.466

Yes 141 0.194

Yes 141 0.282

Yes 282 0.288

Yes 282 0.370

Observed × Contribute Constant

Middle

Demographic controls: Gender (Male/Female), number of undergraduate semesters, and number of undergraduate economics classes. Standard errors clustered at the group level. ∗

9 10 11 12

p < 0.10,

∗∗

p < 0.05,

∗∗∗

p < 0.01

First we consider the results presented in columns (1), (3), and (5). As above, this specification holds the marginal effect of one’s own contribution constant for each endowment across treatments. For each endowment we observe that increasing one’s contribution significantly lowers the amount of punishment received, and an increase in the contributions 21

13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 1 2 3 4 5 6 7 8 9 10 11 12

of one’s group members increases the amount of punishment received. This suggests that group members were sanctioned less as they increased their contributions and that more cooperative groups tended to impose more sanctions. The average marginal effect of Observed varies across endowments. All else equal, a Low member can expect significantly less punishment, a Middle member can expect insignificantly more, and a High member can expect significantly more when endowments are observed relative to when they are unobserved. To investigate how sanctions were distributed, we interact Observed with one’s contribution in columns (2), (4), and (6). First note that an increase in the contributions of one’s group remains positive and significant. The more interesting implication of this specification is to consider the marginal effect of increasing one’s contribution on the amount of sanctions received across treatments. For Low members in Unobserved increasing one’s contribution significantly reduces the amount of punishment received, and the marginal effect is larger and significant in Observed (−1.92 (0.28), p < 0.01). For Middle members in Unobserved increasing one’s contribution has no significant effect on the amount of punishment received. However, there is a significant reduction in Observed and the marginal effect is negative and significant (−0.97 (0.30), p < 0.01). For High members in Unobserved increasing one’s contribution significantly lowers the amount of punishment received. Further, there is a significant reduction in Observed and the marginal effect is negative and significant (−1.18 (0.23), p < 0.01). Overall, the impact of increasing one’s contribution in Observed has a larger, negative, effect on the amount of punishment received for each endowment. The above analysis implies that the benefit, measured as the reduction in punishment received, of increasing one’s contribution is greater when endowments are observed. To further demonstrate this we estimate the predicted difference in punishment received (Observed minus Unobserved ) by one’s contribution using the models presented in (2), (4) and (6) of Table 4 in Figure 6. This analysis largely reinforces the results presented above. Cooperating Low types receive significantly more punishment in Unobserved while freeriding High types receive significantly less punishment in Unobserved. In other words, incomplete information reduces both the benefit of cooperation and the cost of freeriding.

22

Low

Middle

20

10

0

-10

30

Predicted difference in punishment

30

Predicted difference in punishment

Predicted difference in punishment

30

High

20

10

0

-10 0

5

Contribution

10

20

10

0

-10 0

10

Contribution

20

0

10

20

30

Contribution

Figure 6: Average marginal effects for Table 4 by one’s contribution. The vertical axis shows the predicted difference in punishment in Observed relative to Unobserved. The shaded region is the 95 percent confidence interval.

13

14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11

4. Discussion We studied the role of incomplete information in the choice between peer punishment and central authority enforcement institutions designed to induce cooperation and enhance group welfare in social dilemmas. We presented three, linear public good experiments: one in which each group member receives an equal endowment (Equal ), and two in which group members receive unequal endowments and either observe endowments of other group members (Observed ) or not (Unobserved ). Groups were exposed to both a peer punishment and a central authority regime before voting for one enforcement institution to self-impose. When endowments across subjects were unequal, we found that individuals in Observed tended to prefer peer punishment, while individuals in Unobserved, particularly those with lower endowments, tended to prefer central authority. This is largely because those with lower endowments earned less under incomplete information, while high endowment subjects earned about the same in Observed as Unobserved. Since our main research question focused on the cost of incomplete information on decentralized enforcement, we designed our experiment with a simplistic central authority institution to better delineate the choice between the two enforcement institutions. In contrast to Nicklisch et al. (2016), the alternative institution here is a deterrent central authority with perfect information which enforces the social optimum, but at a relatively high fixed cost. As such, we created an alternative which is equally effective across all treatments and 23

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 1 2 3 4 5 6 7 8 9 10 11 12

investigated how the relative effectiveness of peer punishment alters institutional preference across treatments. Our results complement the findings of Nicklisch et al. (2016) and suggest that incomplete information about endowments, or resources more generally, could help explain why central authority institutions have come to dominate modern society. However, it is important to recognize that a realistic central authority will not resemble the all-knowing adjudicator in our design. If individuals can somehow fool or even corrupt the central authority, then there may be conditions under which peer punishment is actually preferred to central authority. Or as in Nicklisch et al. (2016), conditions under no enforcement institution is adopted. Lessons from the mechanism design literature remind us there exist no realistic mechanisms that induce efficient provision of public goods when agents have private information (Hurwicz, 1972). By realistic we mean mechanisms that are incentive compatible, satisfy an individual’s participation constraint, and work for any distribution of preferences. It so happens that the only mechanism that can maximize efficiency under these three conditions is a dictator-style enforcer similar to our central authority (Satterthwaite, 1975). Exactly what is the second-best (and arguably more appealing) central authority is a promising topic for future research. Future work can also explore other informal alternatives to central authority (e.g. ostracism), or allow subjects to vote for a VCM and opt of out enforcement altogether (similar to Markussen et al. (2014)). Examining long-run dynamics of income inequality and incomplete information is another important topic for future work. Since our main interest was the choice of enforcement institutions by groups, our design used short phases, the upside being that subjects gained more experience with each institution and had more opportunities to vote. The downside is that we are unable to tease out the emergent contribution norms that motivated punishment like Carpenter and Matthews (2009) and Reuben and Riedl (2013). In particular, Reuben and Riedl (2013) show that under complete information, subjects enforce multiple contribution norms that account for differences in wealth. However, our results do suggest that targeting is ineffective under incomplete information, which may stunt the emergence of optimal contribution norms. Moreover, subjects may further exploit incomplete information by establishing contribution norms that do not require them to reveal their wealth. Notably, our study does not delve into the nuances of punishment decisions due to the few periods at our disposal. Future work can explore how subjects wield punishment in similar experiments with longer time horizons. Finally, our study points to the importance of understanding how income inequality –

24

13 14 15 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

and more broadly, heterogeneity in resources – affects social capital in groups.27 Our findings suggest that societies faced with less heterogeneity may be less dependent on formal institutions because they more effectively employ informal, peer to peer, institutions (Ostrom, 1990). In our introduction we pointed to evidence from the micro-finance literature that suggests homogeneous groups are better at attracting and repaying loans (Cassar et al., 2007). Group lending is a commonly used tool to deliver credit to the poor in developing countries and is, by definition, informal. Such arrangements depend on informal, peer-topeer mechanisms to encourage repayment of loans that are given without collateral. Other research suggests that trust and social norms are highest in socially homogeneous and low income-inequality countries, while investment in public goods like education is lowest in countries with high income inequality (Knack and Keefer, 1997; Zak and Knack, 2001; Easterly, 2007). Knack and Keefer (1997) show that trust and cooperative social norms are conducive for growth and that these characteristics are higher in socially homogeneous and low income inequality societies. Zak and Knack (2001) model the relationship between trust and economic performance and argue that low trust reduces investment and subsequent growth. This low-trust poverty trap is more likely when existing institutions are weak and society is heterogeneous. Alesina and La Ferrara (2000) investigate the source of social capital with respect to engagement in civic associations and find that civic engagement declines as communities become more heterogeneous and unequal.

27

The term social capital captures the stock of social norms, networks of civic engagement, and the trustworthiness that define a society or group (Coleman, 2000).

25

17

18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

References Alesina, A., La Ferrara, E., 2000. Participation in heterogeneous communities. The Quarterly Journal of Economics 115 (3), 847–904. Ambrus, A., Greiner, B., 2012. Imperfect public monitoring with costly punishment: An experimental study. The American Economic Review 102 (7), 3317–3332. Andreoni, J., Gee, L. K., 2012. Gun for hire: Delegated enforcement and peer punishment in public goods provision. Journal of Public Economics 96 (11-12), 1036–1046. Becker, G. S., 1968. Crime and punishment: An economic approach. In: The economic dimensions of crime. Springer, pp. 13–68. Bochet, O., Page, T., Putterman, L., 2006. Communication and punishment in voluntary contribution experiments. Journal of Economic Behavior & Organization 60 (1), 11–26. Buckley, E., Croson, R., 2006. Income and wealth heterogeneity in the voluntary provision of linear public goods. Journal of Public Economics 90, 935–955. Carpenter, J., Matthews, P. H., 2009. What norms trigger punishment? Experimental Economics 12 (3), 272–288. Cassar, A., Crowley, L., Wydick, B., 2007. The effect of social capital on group loan repayment: Evidence from field experiments. The Economic Journal 117 (517). Chaudhuri, A., 2011. Sustaining cooperation in laboratory public goods experiments: A selective survey of the literature. Experimental Economics 14 (1), 47–83. Coleman, J. S., 2000. Social capital in the creation of human capital. In: Knowledge and social capital. Elsevier, pp. 17–41. Davis, D. D., Holt, C. A., 1993. Experimental Economics. Princeton University Press. Easterly, W., 2007. Inequality does cause underdevelopment: Insights from a new instrument. Journal of Development Economics 84 (2), 755–776. Egas, M., Riedl, A., 2008. The economics of altruistic punishment and the maintenance of cooperation. Proceedings of the Royal Society of London B: Biological Sciences 275 (1637), 871–878. Ehrlich, I., 1996. Crime, punishment, and the market for offenses. The Journal of Economic Perspectives 10 (1), 43–67. Ertan, A., Page, T., Putterman, L., 2009. Who to punish? individual decisions and majority rule in mitigating the free rider problem. European Economic Review 53 (5), 495–511. Fehr, E., Gächter, S., 2000. Cooperation and punishment in public goods experiments. The American Economic Review 90 (4), 980–994. Fehr, E., Gächter, S., 2002. Altruistic punishment in humans. Nature 415 (6868), 137–140. Fellner, G., Iida, Y., Kröger, S., Seki, E., 2011. Heterogeneous productivity in voluntary public good provision: an experimental analysis. IZA DP No. 5556. Fischbacher, U., 2007. z-tree: Zurich toolbox for ready-made economic experiments. Experimental Economics 10 (2), 171–178. Gächter, S., Herrmann, B., 2008. Reciprocity, culture and human cooperation: Previous insights and a new cross-cultural experiment. Philosophical Transactions of the Royal Society B: Biological Sciences. Gächter, S., Renner, E., Sefton, M., 2008. The long-run benefits of punishment. Science 322 (5907), 1510– 1510.

26

17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 1 2 3 4 5 6 7 8 9 10 11 662 663 664 665 666

Galbiati, R., Vertova, P., 2008. Obligations and cooperative behaviour in public good games. Games and Economic Behavior 64 (1), 146–170. Grechenig, K., Nicklisch, A., Thöni, C., 2010. Punishment despite reasonable doubt? a public goods experiment with sanctions under uncertainty. Journal of Empirical Legal Studies 7 (4), 847–867. Gurerk, O., Irlenbusch, B., Rockenbach, B., 2006. The competitive advantage of sanctioning institutions. Science 312, 108–110. Harsanyi, J. C., 1967. Games with incomplete information played by “bayesian” players, i–iii part i. the basic model. Management Science 14 (3), 159–182. Hurwicz, L., 1972. On informationally decentralized systems. Decision and Organization. Kamei, K., Putterman, L., Tyran, J.-R., 2015. State or nature? endogenous formal versus informal sanctions in the voluntary provision of public goods. Experimental Economics 18 (1), 38–65. Kamijo, Y., Nihonsugi, T., Takeuchi, A., Funaki, Y., 2014. Sustaining cooperation in social dilemmas: Comparison of centralized punishment institutions. Games and Economic Behavior 84, 180–195. Kingsley, D. C., 2016. Endowment heterogeneity and peer punishment in a public good experiment: Cooperation and normative conflict. Journal of Behavioral and Experimental Economics 60, 49–61. Knack, S., Keefer, P., 1997. Does social capital have an economic payoff? a cross-country investigation. The Quarterly Journal of Economics 112 (4), 1251–1288. Kosfeld, M., Okada, A., Riedl, A., 2009. Institution formation in public goods games. American Economic Review 99 (4), 1335–1355. Ledyard, J. O., 1995. Public goods: A survey of experimental research. s. 111–194 in: J. kagel/a. roth (hrsg.), handbook of experimental economics. Markussen, T., Putterman, L., Tyran, J.-R., 2014. Self-organization for collective action: An experimental study of voting on sanction regimes. Review of Economic Studies 81, 301–324. Neitzel, J., Saaksvuori, L., 2013. Normative conflict and cooperation in sequential social dilemmas. Annual Conference 2013 (Duesseldorf: Competition Policy and Regulation in a Global Economic Order). Nicklisch, A., Grechenig, K., Thöni, C., 2016. Information-sensitive leviathans. Journal of Public Economics 144, 1–13. Nikiforakis, N., 2008. Punishment and counter-punishment in public good games: Can we really govern ourselves? Journal of Public Economics 92 (1-2), 91–112. Nikiforakis, N., Normann, H.-T., 2008. A comparative statics analysis of punishment in public-good experiments. Experimental Economics 11 (4), 358–369. Nikiforakis, N., Noussair, C. N., Wilkening, T., 2012. Normative conflict and feuds: The limits of selfenforcement. Journal of Public Economics 96 (9-10), 797–807. Ostrom, E., 1990. Governing the Commons. Cambridge University Press. Ostrom, E., Walker, J., Gardner, R., 1992. Covenants with and without a sword: Self-governance is possible. American Political Science Review 86 (02), 404–417. Polinsky, A. M., Shavell, S., 1979. The optimal tradeoff between the probability and magnitude of fines. The American Economic Review 69 (5), 880–891. Putterman, L., Tyran, J.-R., Kamei, K., 2011. Public goods and voting on formal sanction schemes. Journal of Public Economics 95 (9-10), 1213–1222. Reuben, E., Riedl, A., 2013. Enforcement of contribution norms in public good games with heterogeneous

27

667 668 669 670 671 672 673 674 675 676 677

populations. Games and Economic Behavior 77 (1), 122–137. Satterthwaite, M. A., 1975. Strategy-proofness and arrow’s conditions: Existence and correspondence theorems for voting procedures and social welfare functions. Journal of Economic Theory 10 (2), 187–217. Sefton, M., Shupp, R., Walker, J. M., 2007. The effect of rewards and sanctions in provision of public goods. Economic Inquiry 45 (4), 671–690. Stigler, G. J., 1970. The optimum enforcement of laws. Journal of Political Economy 78 (3), 526–536. Sutter, M., Haigner, M., G., K., 2010. Choosing the carrot or the stick? endogenous institutional choice in social dilemma situations. The Review of Economic Studies 77 (4), 1540–1566. Tyran, J.-R., Feld, L. P., 2006. Achieving compliance when legal sanctions are non-deterrent. The Scandinavian Journal of Economics 108 (1), 135–156. Zak, P. J., Knack, S., 2001. Trust and growth. The Economic Journal 111 (470), 295–321.

28